more accurate forecast data

Question

I live in part of the world, wehre the weather forecast as provided by WU is sometimes not accurate enough, e.g. southerly winds like foehn and its consequences are poorly taken into account by WU. Unfortunately, the local weather bureau is a closed fish tank, not providing any API to the public, hence the idea of parsing their icons of clouds, rain, and sun every 12 hours or so and feeding this information into the nice WeahterStation products by ThingPulse.

Has anyone a clue, how to program this parser and enhance the WeatherStation w.r.t. its forecast accuracy?

Comments

We were force to migrate all our code to OpenWeatherMap earlier this year, see https://thingpulse.com/hello-openweathermap-bye-bye-wunderground/ for details. The WU code is deprecated and will be removed next June. Did you check whether the OWM data (e.g. https://openweathermap.org/city/2660594) is any more useful than what WU provides?

---

Thanks for your quick reply. I managed to migrate to OpenWeatherMap an I'll put my sceptical eyes on the weather station's display and watch the real world though the window.

---

You may find it useful to look at the Zambretti forecaster implemented in ESP code, it uses only air pressure variations and can optionally use temperature to give an accuracy of approaching 90% or more. Example: https://github.com/G6EJD/ESP32_Weather_Forecaster_TN061/blob/master/ESP32_BMP180_Forecaster_29_epaper_v2b.ino

All you need to do it parse current pressure and its trend ("+","0" or "-") to calc_zambretti(float zpressure, String ztrend) and it will return a forecast in the variable wx_text and an image-type (e.g. sunny, cloudy, rain) in wx_image

---

Thanks for your hint - although, I have an old school barometer next to my ThingPuls Color Weather Station. Luckily, I have an BMP280 at hand, but  no 2.9" ePaper, so it'll take a bit of coding for me to implement the Zambetti algorithm into the carousel...

---

You can also get current air pressure change from WU and use their trend indicator of -, 0 or + to denote falling, steady and rising, so no need for an external BME280. I used the same nomenclature as WU for trend to make it easier. Zambretti forecasts are used worldwide for a reliable and they say >90% accurate forecast and my experience shows that to be the case, although I have not run an experiment of actual v forecast weather to get a matric.

---

Well, I am insisting of NOT using WU or OW data, but rather acquire local meteo data: in a chart of maximum daylight temperatures of OW versus our local meteo forecast for the next five days gives me a correlation factor of 0.02 (yes, we have a classical foehn situation reaching 8 hPa southerly overpressure here ;-) which is not a satisfying forecast of OW at all. And even the Zambretti would underestimate the effect of the approach of the next low-pressure on the maximum daylight temperature.

Answer 1

The Zambetti forecast is most accurate at just after sunrise, so if you set the pressure and trend and optionally wind direction it will provide a 94% accurate forecast for at least 24-hours.

Compared with WU and the UK Meto I find forecasts from OW to be similarly inaccurate. Today in the UK are clear blue skies but OW says it’s going to rain, clearly it is not so their forecast model needs some more work.

Creating your own forecast model is not easy but in one of my tech notes I included a set of forecasting rules used by NOAA for the northern hemisphere and they are like the Zambetti imperical and some what qualitative but nonetheless reasonably accurate. There are about 20 rules using pressure, pressures trend, wind direction and temperature so you have the raw data from say OW or WU to apply to the rules.